# -*- coding: utf-8 -*-
import json, io, sys, re
sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding='utf-8')
S = json.load(open('_qa_slides.json', encoding='utf-8'))

ROMAN = ["Ⅰ","Ⅱ","Ⅲ","Ⅳ","Ⅴ","Ⅵ","Ⅶ","Ⅷ"]

print("=== PER-SLIDE STRUCTURE (chapter line + first content) ===")
for s in S:
    n = s["n"]
    chap = ""
    for t in s["texts"]:
        # chapter header like "Ⅲ. 기술·기능 · ..."
        if any(t.startswith(r + ".") or t.startswith(r + " ") for r in ROMAN):
            chap = t
            break
    # headline = the longest early text that's not chapter/boilerplate
    head = ""
    for t in s["texts"][:6]:
        if t == chap: continue
        if "킨텍스" in t or "2026-07" in t or t in ROMAN: continue
        if len(t) > len(head): head = t
    print(f"{n:>3}| {chap[:46]:<46} | {head[:52]}")

print("\n=== 배점 / SCORING TOKENS ===")
pats = ["80/20","35/80","90/10","70+20","90(","기술 90","기술 80","가격 10","가격 20","정성 70","정량 20","재정렬","공고 확정"]
for p in pats:
    hits=[(s['n'],t[:80]) for s in S for t in s['texts'] if p in t]
    print(f'--- {p!r}: {len(hits)}')
    for n,t in hits[:8]: print(f'    {n}: {t}')

print("\n=== HONESTY LABELS ===")
for p in ["[검증됨]","[구현 계획]","[협의]","[가정]","[Non-Goal]","[공고"]:
    hits=[(s['n'],t[:74]) for s in S for t in s['texts'] if p in t]
    print(f'--- {p!r}: {len(hits)}')
    for n,t in hits[:14]: print(f'    {n}: {t}')

print("\n=== WATERMARK / AI NOTICE ===")
for p in ["워터마크","AI 생성","AI-generated","watermark","합성 이미지","예상 이미지","시공 예"]:
    hits=[(s['n'],t[:74]) for s in S for t in s['texts'] if p.lower() in t.lower()]
    print(f'--- {p!r}: {len(hits)} : slides {sorted(set(n for n,_ in hits))}')